The most common functions and classes are available in the
gevent top level package.
The human-readable PEP 440 version identifier
version_info= version_info(major=1, minor=1, micro=0, releaselevel='rc', serial='4')¶
The programatic version identifier. The fields have (roughly) the
same meaning as
Greenlet is a light-weight cooperatively-scheduled execution unit.
To start a new greenlet, pass the target function and its arguments to
Greenlet constructor and call
>>> g = Greenlet(myfunction, 'arg1', 'arg2', kwarg1=1) >>> g.start()
or use classmethod
spawn() which is a shortcut that does the same:
>>> g = Greenlet.spawn(myfunction, 'arg1', 'arg2', kwarg1=1)
To subclass a
Greenlet, override its
_run() method and
__init__(): It also a good
idea to override
_run() raises an exception,
its string representation will be printed after the traceback it
You SHOULD NOT attempt to override the
__init__(run=None, *args, **kwargs)¶
Changed in version 1.1a3: The
run argument to the constructor is now verified to be a callable
object. Previously, passing a non-callable object would fail after the greenlet
Holds the value returned by the function if the greenlet has
finished successfully. Until then, or if it finished in error,
Recall that a greenlet killed with the default
GreenletExit is considered to have finished
successfully, and the
GreenletExit exception will be
Holds the exception instance raised by the function if the greenlet has finished with an error.
Return a true value if and only if the greenlet has finished execution.
Changed in version 1.1: This function is only guaranteed to return true or false values, not
necessarily the literal constants
Return a true value if and only if the greenlet has finished execution successfully, that is, without raising an error.
A greenlet that has been killed with the default
GreenletExit exception is considered successful.
GreenletExit is not considered an error.
This function is only guaranteed to return true or false values,
not necessarily the literal constants
Schedule the greenlet to run in this loop iteration
Schedule the greenlet to run in the future loop iteration seconds later
Wait until the greenlet finishes or timeout expires.
Return the result the greenlet has returned or re-raise the exception it has raised.
If block is
gevent.Timeout if the greenlet is still alive.
If block is
True, unschedule the current greenlet until the result is available
or the timeout expires. In the latter case,
gevent.Timeout is raised.
kill(exception=GreenletExit, block=True, timeout=None)¶
exception in the greenlet.
True (the default), wait until the greenlet dies or the optional timeout expires.
If block is
False, the current greenlet is not unscheduled.
The function always returns
None and never raises an error.
Depending on what this greenlet is executing and the state
of the event loop, the exception may or may not be raised
immediately when this greenlet resumes execution. It may
be raised on a subsequent green call, or, if this greenlet
exits before making such a call, it may not be raised at
all. As of 1.1, an example where the exception is raised
later is if this greenlet had called
sleep(0); an example where the exception is raised
immediately is if this greenlet had called
|Parameters:||exception (type) – The type of exception to raise in the greenlet. The default
Changed in version 0.13.0: block is now
True by default.
Changed in version 1.1a2: If this greenlet had never been switched to, killing it will prevent it from ever being switched to.
Link greenlet’s completion to a callable.
The callback will be called with this instance as an argument once this greenlet’s dead. A callable is called in its own greenlet.
link() but callback is only notified when the greenlet has completed successfully.
link() but callback is only notified when the greenlet dies because of an unhandled exception.
Register a callable to be executed when the greenlet finishes execution.
The callback will be called with this instance as an argument.
The callable will be called in the HUB greenlet.
Greenlet objects have a boolean value (
__bool__) which is true if it’s active: started but not dead yet.
It’s possible to use it like this:
>>> g = gevent.spawn(...) >>> while g: # do something while g is alive
__nonzero__ is an improvement on greenlet’s
__nonzero__. The greenlet’s
__nonzero__ returns False if greenlet has not
been switched to yet or is already dead. While the latter is OK, the
former is not good, because a just spawned Greenlet has not been
switched to yet and thus would evaluate to False.
Being a greenlet subclass,
Greenlet also has switch() and throw() methods. However, these should
not be used at the application level as they can very easily lead to
greenlets that are forever unscheduled. Prefer higher-level safe
A special exception that kills the greenlet silently.
When a greenlet raises
GreenletExit or a subclass, the traceback is not
printed and the greenlet is considered
The exception instance is available under
property as if it was returned by the greenlet, not raised.
spawn(function, *args, **kwargs)¶
Create a new
Greenlet object and schedule it to run
This can be used as
The arguments are passed to
spawn_later(seconds, function, *args, **kwargs)¶
Create and return a new Greenlet object scheduled to run
in the future loop iteration seconds later. This can be used as
The arguments are passed to
Changed in version 1.1a3: If an argument that’s meant to be a function (the first argument in args, or the
run keyword )
is given to this classmethod (and not a classmethod of a subclass),
it is verified to be callable. Previously, the spawned greenlet would have failed
when it started running.
spawn_raw(function, *args, **kwargs)¶
Create a new
greenlet.greenlet object and schedule it to
This returns a raw
greenlet which does not have all the useful
gevent.Greenlet has. Typically, applications
spawn(), but this method may
occasionally be useful as an optimization if there are many
Changed in version 1.1rc2: Accept keyword arguments for
function as previously (incorrectly)
documented. Note that this may incur an additional expense.
Changed in version 1.1a3: Verify that
function is callable, raising a TypeError if not. Previously,
the spawned greenlet would have failed the first time it was switched to.
Return the currently executing greenlet (the one that called this
function). Note that this may be an instance of
Put the current greenlet to sleep for at least seconds.
seconds may be specified as an integer, or a float if fractional seconds are desired.
In the current implementation, a value of 0 (the default) means to yield execution to any other runnable greenlets, but this greenlet may be scheduled again before the event loop cycles (in an extreme case, a greenlet that repeatedly sleeps with 0 can prevent greenlets that are ready to do I/O from being scheduled for some (small) period of time); a value greater than 0, on the other hand, will delay running this greenlet until the next iteration of the loop.
If ref is False, the greenlet running
sleep() will not prevent
Cause the calling greenlet to wait until the event loop is idle.
Idle is defined as having no other events of the same or higher priority pending. That is, as long as sockets, timeouts or even signals of the same or higher priority are being processed, the loop is not idle.
Kill greenlet asynchronously. The current greenlet is not unscheduled.
Greenlet.kill() method does the same and
more (and the same caveats listed there apply here). However, the MAIN
greenlet - the one that exists initially - does not have a
kill() method, and neither do any created with
so you have to use this function.
Changed in version 1.1a2: If the
greenlet has a
kill method, calls it. This prevents a
greenlet from being switched to for the first time after it’s been
killed but not yet executed.
killall(greenlets, exception=GreenletExit, block=True, timeout=None)¶
Forceably terminate all the
greenlets by causing them to raise
Timeout – If blocking and a timeout is given that elapses before all the greenlets are dead.
wait(objects=None, timeout=None, count=None)¶
objects to become ready or for event loop to finish.
objects is provided, it must be a list containing objects
implementing the wait protocol (rawlink() and unlink() methods):
None (the default),
wait() blocks until
the current event loop has nothing to do (or until
None (the default), wait for all
to become ready.
count is a number, wait for (up to)
count objects to become
ready. (For example, if count is
1 then the function exits
when any object in the list is ready).
timeout is provided, it specifies the maximum number of
wait() will block.
Returns the list of ready objects, in the order in which they were ready.
iwait(objects, timeout=None, count=None)¶
Iteratively yield objects as they are ready, until all (or count) are ready or timeout expired.
joinall(greenlets, timeout=None, raise_error=False, count=None)¶
Wait for the
greenlets to finish.
A sequence of the greenlets that finished before the timeout (if any) expired.
Forks a child process and starts a child watcher for it in the
parent process so that
waitpid and SIGCHLD work as expected.
This implementation of
fork is a wrapper for
when the environment variable
GEVENT_NOWAITPID is not defined.
This is the default and should be used by most applications.
Changed in version 1.1b2.
Prepare the gevent hub to run in a new (forked) process.
This should be called immediately after
os.fork() in the
child process. This is done automatically by
gevent.os.fork() or if the
os module has been
monkey-patched. If this function is not called in a forked
process, symptoms may include hanging of functions like
socket.getaddrinfo(), and the hub’s threadpool is unlikely
Registered fork watchers may or may not run before
this function (and thus
gevent.os.fork) return. If they have
not run, they will run “soon”, after an iteration of the event loop.
You can force this by inserting a few small (but non-zero) calls to
after fork returns. (As of gevent 1.1 and before, fork watchers will
not have run, but this may change in the future.)
This function may be removed in a future major release if the fork process can be more smoothly managed.
See remarks in
gevent.os.fork() about greenlets
and libev watchers in the child process.
signal(signalnum, handler, *args, **kwargs)¶
Call the handler with the args and kwargs when the process receives the signal signalnum.
The handler will be run in a new greenlet when the signal is delivered.
This returns an object with the useful method
cancel, which, when called,
will prevent future deliveries of signalnum from calling handler.
This may not operate correctly with SIGCHLD if libev child watchers
are used (as they are by default with
Changed in version 1.1b4: This is an alias for
gevent.hub.signal, included for
backwards compatibility; the new module gevent.signal
is replacing this name. This alias will be removed in a
Timeout(seconds=None, exception=None, ref=True, priority=-1)¶
Raise exception in the current greenlet after given time period:
timeout = Timeout(seconds, exception) timeout.start() try: ... # exception will be raised here, after *seconds* passed since start() call finally: timeout.cancel()
If the code that the timeout was protecting finishes
executing before the timeout elapses, be sure to
timeout so it is not unexpectedly raised in the future. Even if
it is raised, it is a best practice to cancel it. This
try/finally construct or a
with statement is a
When exception is omitted or
Timeout instance itself is raised:
>>> import gevent >>> gevent.Timeout(0.1).start() >>> gevent.sleep(0.2) Traceback (most recent call last): ... Timeout: 0.1 seconds
To simplify starting and canceling timeouts, the
with statement can be used:
with gevent.Timeout(seconds, exception) as timeout: pass # ... code block ...
This is equivalent to the try/finally block above with one additional feature:
if exception is the literal
False, the timeout is still raised, but the context manager
suppresses it, so the code outside the with-block won’t see it.
This is handy for adding a timeout to the functions that don’t support a timeout parameter themselves:
data = None with gevent.Timeout(5, False): data = mysock.makefile().readline() if data is None: ... # 5 seconds passed without reading a line else: ... # a line was read within 5 seconds
readline() above catches and doesn’t re-raise
(for example, with a bare
except:), then your timeout will fail to function and control
won’t be returned to you when you expect.
When catching timeouts, keep in mind that the one you catch may not be the one you have set (a calling function may have set its own timeout); if you going to silence a timeout, always check that it’s the instance you need:
timeout = Timeout(1) timeout.start() try: ... except Timeout as t: if t is not timeout: raise # not my timeout
If the seconds argument is not given or is
Timeout()), then the timeout will never expire and never raise
exception. This is convenient for creating functions which take
an optional timeout parameter of their own.
Changed in version 1.1b2: If seconds is not given or is
None, no longer allocate a libev
timer that will never be started.
If the timeout is pending, cancel it. Otherwise, do nothing.
Return True if the timeout is scheduled to be raised.
Schedule the timeout.
start_new(timeout=None, exception=None, ref=True)¶
Create a started
This is a shortcut, the exact action depends on timeout‘s type:
Timeout, then call its
start()method if it’s not already begun.
Timeoutinstance, passing (timeout, exception) as arguments, then call its
with_timeout(seconds, function, *args, **kwds)¶
Wrap a call to function with a timeout; if the called function fails to return before the timeout, cancel it and return a flag value, provided by timeout_value keyword argument.
If timeout expires but timeout_value is not provided, raise
Keyword argument timeout_value is not passed to function.
Next page: Networking interfaces